The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the belt command-line utility to interface with the Pruna P-Video-Avatar and P-Image models. The allowed-tools frontmatter correctly limits command execution to the belt tool, reducing the risk of arbitrary command execution.
[EXTERNAL_DOWNLOADS]: The instructions reference a remote installation guide (cli-install.md) and a Node.js package (belt-sh/cli) associated with the inference-sh platform. These are documented as functional dependencies for the integrated service.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external inputs such as user-provided text scripts and remote media URLs, which represents a potential injection surface.
Ingestion points: The voice_script, image, and audio parameters in SKILL.md accept external content intended for video generation.
Boundary markers: External inputs are encapsulated within JSON payloads passed to the CLI tool.
Capability inventory: The skill has access to shell execution via the restricted belt command.
Sanitization: The instructions do not specify explicit sanitization steps, as filtering is expected to be handled by the underlying service.

p-video-avatar